On Thu, 2017-08-31 at 18:18 +0200, Sebastian Andrzej Siewior wrote:
> On 2017-08-23 11:53:44 [+0200], Mike Galbraith wrote:
> > virt box reminded me this morning to report this gripe.
>
> if you can reproduce it, then this should make it go away:
Missed a spot. With this on top, lockdep went silent.
kernel/hrtimer/hotplug: don't wake ktimersoftd while holding the hrtimer base lock
kernel/hrtimer: don't wakeup a process while holding the hrtimer base lock
missed a path, namely hrtimers_dead_cpu() -> migrate_hrtimer_list(). Defer
raising softirq until after base lock has been released there as well.
Signed-off-by: Mike Galbraith <efault@xxxxxx>
---
kernel/time/hrtimer.c | 19 +++++++++++++------
1 file changed, 13 insertions(+), 6 deletions(-)
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1779,7 +1779,7 @@ int hrtimers_prepare_cpu(unsigned int cp
#ifdef CONFIG_HOTPLUG_CPU
-static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
+static int migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
struct hrtimer_clock_base *new_base)
{
struct hrtimer *timer;
@@ -1809,15 +1809,19 @@ static void migrate_hrtimer_list(struct
}
#ifdef CONFIG_PREEMPT_RT_BASE
list_splice_tail(&old_base->expired, &new_base->expired);
- if (!list_empty(&new_base->expired))
- raise_softirq_irqoff(HRTIMER_SOFTIRQ);
+ /*
+ * Tell the caller to raise HRTIMER_SOFTIRQ. We can't safely
+ * acquire ktimersoftd->pi_lock while the base lock is held.
+ */
+ return !list_empty(&new_base->expired);
#endif
+ return 0;
}
int hrtimers_dead_cpu(unsigned int scpu)
{
struct hrtimer_cpu_base *old_base, *new_base;
- int i;
+ int i, raise = 0;
BUG_ON(cpu_online(scpu));
tick_cancel_sched_timer(scpu);
@@ -1833,13 +1837,16 @@ int hrtimers_dead_cpu(unsigned int scpu)
raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
- migrate_hrtimer_list(&old_base->clock_base[i],
- &new_base->clock_base[i]);
+ raise |= migrate_hrtimer_list(&old_base->clock_base[i],
+ &new_base->clock_base[i]);
}
raw_spin_unlock(&old_base->lock);
raw_spin_unlock(&new_base->lock);
+ if (raise)
+ raise_softirq_irqoff(HRTIMER_SOFTIRQ);
+
/* Check, if we got expired work to do */
__hrtimer_peek_ahead_timers();
local_irq_enable();
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
